Re: Binary support for pgoutput plugin

I've pushed this patch, with a number of adjustments, some cosmetic
and some not so much (no pg_dump support!?). We're not quite
done though ...

Dave Cramer <davecramer(at)gmail(dot)com> writes:
> So looking at how to confirm that the subscriber has receive functions for
> all of the types.
> AFAICT we don't have that information since the publication determines what
> is sent?

Yeah, at the point where we need to send the option, we seem not to have a
lot of info. In practice, if the sender has a typsend function, the only
way the subscriber doesn't have a matching typreceive function is if it's
an older PG version. I think it's sufficient to document that you can't
use binary mode in that case, so that's what I did. (Note that
getTypeBinaryInputInfo will say "no binary input function available for
type %s" in such a case, so that seemed like adequate error handling.)

> On Tue, 14 Jul 2020 at 22:47, Andres Freund <andres(at)anarazel(dot)de> wrote:
>> An oid mismatch error without knowing what that's about isn't very
>> helpful either.
>> How about adding an errcontext that shows the "source type oid", the
>> target type oid & type name and, for records, the column name of the
>> target table? That'd make this a lot easier to debug.

> This code line 482 in proto.c attempts to limit what is sent in binary. We
> could certainly be more restrictive here.

I think Andres' point is to be *less* restrictive. I left that logic
as-is in the committed patch, but we could do something like the attached
to improve the situation.

regards, tom lane